Asset evaluation method based on multi-agent cooperation and related device

Abstract

The application discloses an asset evaluation method based on multi-agent cooperation and related devices, the method comprises the following steps: training a preset asset evaluation model by using data in a first preset database to obtain a target asset evaluation model; the target asset evaluation model comprises a target base model and a plurality of professional agents; the plurality of professional agents correspond to a plurality of professional tasks; each professional agent corresponds to a professional task; a is an integer greater than 1; an asset evaluation instruction of a target object is obtained; target asset data to be evaluated is determined according to the asset evaluation instruction; the target asset data to be evaluated is processed by the target base model to obtain first evaluation data and target data evaluation requirements; the first evaluation data is processed by the plurality of professional agents according to the target data evaluation requirements to obtain a target asset evaluation result. The application improves the asset evaluation efficiency.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to an asset evaluation method and related apparatus based on multi-agent collaboration. Background Technology

[0002] With the rapid development of big data and artificial intelligence technologies, the general large language model (LLM) has been widely used by various institutions and enterprises, which has greatly promoted the marketization of data elements. The valuation of data assets has become a core and key step in the circulation of data elements, such as data ownership confirmation, value pricing, asset trading and accounting entry. The scientificity, accuracy and efficiency of its valuation results directly affect the market allocation efficiency of data elements.

[0003] Currently, data asset assessment relies heavily on the industry experience of human experts. The assessment process is greatly influenced by the subjective judgment of experts, the assessment results lack unified quantitative standards, and the assessment time is long, resulting in low assessment efficiency.

[0004] Therefore, how to improve the efficiency of asset valuation has become an urgent problem to be solved. Summary of the Invention

[0005] This application provides an asset appraisal method and related apparatus based on multi-agent collaboration, which improves the efficiency of asset appraisal.

[0006] In a first aspect, embodiments of this application provide an asset valuation method based on multi-agent collaboration, comprising: The target asset valuation model is obtained by training a preset asset valuation model using data from the first preset database. The target asset valuation model includes: a target base model and a professional intelligent agents. The a professional intelligent agents correspond to a professional tasks. Each professional intelligent agent corresponds to one professional task. a is an integer greater than 1. Obtain asset valuation instructions for the target entity; The target asset data to be appraised is determined according to the asset appraisal instruction; The target asset data to be evaluated is processed using the target base model to obtain the first evaluation data and the target data evaluation requirements. The a specialized intelligent agents process the first evaluation data according to the target data evaluation requirements to obtain the target asset evaluation result.

[0007] Secondly, embodiments of this application provide an asset evaluation device based on multi-agent collaboration, the device comprising: a training module, an acquisition module, and an asset evaluation module, wherein: The training module is used to train a preset asset valuation model using data from a first preset database to obtain a target asset valuation model. The target asset valuation model includes: a target base model and a professional intelligent agents. The a professional intelligent agents correspond to a professional tasks. Each professional intelligent agent corresponds to one professional task. a is an integer greater than 1. The acquisition module is used to acquire the asset valuation instructions of the target object; The asset valuation module is used to determine the target asset data to be valued according to the asset valuation instruction; process the target asset data to be valued through the target base model to obtain the first valuation data and the target data valuation requirements; and process the first valuation data according to the target data valuation requirements through the a professional intelligent agents to obtain the target asset valuation result.

[0008] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for performing the steps in the first aspect of embodiments of this application.

[0009] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of embodiments of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of embodiments of this application. The computer program product may be a software installation package.

[0011] Implementing this application will have the following beneficial effects: As can be seen, the asset appraisal method based on multi-agent collaboration described in this application obtains a target asset appraisal model through model training. This target asset appraisal model includes a target base model and a professional intelligent agents. The target base model quickly completes the parsing of the asset data to be appraised and the decomposition of appraisal requirements. Then, the a professional intelligent agents perform professional and parallel processing on the first appraisal data according to the appraisal requirements. This realizes the full-process automation and collaboration from data processing and requirement analysis to multi-dimensional appraisal, replacing the inefficient links such as experience dependence, serial processing, and report writing in traditional manual appraisal, thereby improving the efficiency of asset appraisal. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the background art, the accompanying drawings used in the embodiments of this application or the background art will be described below.

[0013] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating an application scenario of an electronic device provided in an embodiment of this application; Figure 3 This is a flowchart of an asset evaluation method based on multi-agent collaboration provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a preset asset valuation model provided in an embodiment of this application; Figure 5 This is a flowchart illustrating the generation process of a data asset valuation report provided in an embodiment of this application; Figure 6 This is a flowchart of a data asset assessment provided in an embodiment of this application; Figure 7 This is a functional module block diagram of an asset appraisal device based on multi-agent collaboration provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of another electronic device provided in an embodiment of this application. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0015] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0016] It should be understood that the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document indicates that the preceding and following related objects are in an "or" relationship. In the embodiments of this application, "multiple" refers to two or more.

[0017] In the embodiments of this application, "at least one item" or its similar expression refers to any combination of these items, including any combination of a single item or a plurality of items. "One or more" means one or more, while "multiple" means two or more. For example, "at least one item" of a, b, or c can represent the following seven cases: a, b, c; a and b; a and c; b and c; a, b, and c. Each of a, b, and c can be an element or a set containing one or more elements.

[0018] In this application embodiment, "connection" refers to various connection methods such as direct connection or indirect connection to realize communication between devices. This application embodiment does not limit this in any way.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] The electronic devices described in this application embodiment may include smartphones (such as Android phones, iOS phones, Windows Phones, etc.), tablet computers, PDAs, laptops, video matrices, monitoring platforms, mobile internet devices (MIDs), or wearable devices, etc. The above are merely examples and not exhaustive, and include but are not limited to the above devices.

[0021] Of course, the aforementioned electronic devices can also be servers, such as cloud servers.

[0022] The following describes the relevant content, concepts, meanings, technical issues, technical solutions, and beneficial effects involved in the embodiments of this application.

[0023] First, let me explain some of the technical terms or phrases used in this application: Asset appraisal: refers to the professional activity of analyzing, evaluating and estimating the value, quality, compliance and other attributes of an asset at a specific point in time, based on relevant standards, criteria and methods. Its core purpose is to provide objective and quantitative reference for asset ownership confirmation, pricing, transaction and inclusion in balance sheets.

[0024] Intelligent agent: refers to an intelligent unit with independent perception, reasoning, decision-making and execution capabilities, which can undertake specific tasks and output corresponding results based on preset rules or learned capabilities.

[0025] General pre-trained large language model: refers to a large language model pre-trained based on massive general text data. It has basic semantic understanding, logical reasoning and text generation capabilities. It can be fine-tuned to adapt to specific scenarios through domain and is the basic carrier for building professional models and intelligent agents.

[0026] RAG technology, or Retrieval Enhanced Generative Technology, refers to retrieving relevant information from an authoritative external knowledge base before the model generates content. This information serves as a contextual constraint, guiding the model to generate results based on the accurate information retrieved. This effectively improves the accuracy and professionalism of the output and reduces the risk of illusions.

[0027] Supervised fine-tuning (SFT): also known as instruction fine-tuning, refers to further training and adjusting a pre-trained language model using labeled task-specific data.

[0028] Contrastive learning is an unsupervised or self-supervised learning method based on the similarity and differences between samples. The core idea of ​​contrastive learning is to learn effective representations of data by comparing the similarities and differences between samples. It does not require a large amount of labeled data; instead, it utilizes the latent relationships between samples (e.g., samples of the same category are more similar) to guide model training. This approach is particularly suitable for scenarios with abundant data but scarce labels, such as sentence representation learning in natural language processing or image retrieval in computer vision.

[0029] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; it can be seen that the electronic device may include: a communication unit, a model training unit, an asset evaluation unit, etc., which are not limited here, wherein: The communication unit is responsible for data interaction between electronic devices and external systems (e.g., the target enterprise system), devices, or terminals. This includes receiving asset assessment instructions, uploading assessment results, synchronizing preset databases and model parameters, and transmitting data of assets to be assessed. It ensures the stability, real-time performance, and security of data transmission and provides data interaction support for processes such as model training and asset assessment.

[0030] The model training unit is used to train, fine-tune, and optimize the preset asset valuation model based on the data in the first preset database, and generate a target asset valuation model that includes a target base model and a professional intelligent agents. At the same time, it completes the model weight solidification, parameter configuration, effect verification, and other tasks to ensure that the model has professional reasoning and evaluation capabilities in the field of data assets.

[0031] The asset appraisal unit is used to call the target base model to complete the parsing of the asset data to be appraised and the breakdown of appraisal requirements. It schedules a professional intelligent agent to perform professional and parallel processing on the first appraisal data according to the appraisal requirements, integrates the output results of each intelligent agent, generates standardized target asset appraisal results, and realizes full-process automated appraisal.

[0032] Please see Figure 2 , Figure 2 This is a schematic diagram illustrating an application scenario of an electronic device provided in an embodiment of this application; it can be seen that, Figure 2 The double-headed arrows indicate that the electronic device can communicate or physically connect with the target object. The electronic device displays two options: "Get Enterprise Instructions" and "Execute Asset Appraisal." Users can click these options to control the electronic device to perform related operations. "Get Enterprise Instructions" means that the electronic device can receive enterprise instructions from the target object; "Execute Asset Appraisal" means that the electronic device can perform an asset appraisal of the target object's assets. Specifically, the electronic device can execute the asset appraisal method based on multi-agent collaboration provided in this application embodiment to perform an asset appraisal. The specific steps are as follows: The target asset valuation model is obtained by training a preset asset valuation model using data from the first preset database. The target asset valuation model includes: a target base model and a professional intelligent agents. The a professional intelligent agents correspond to a professional tasks. Each professional intelligent agent corresponds to one professional task. a is an integer greater than 1. Obtain asset valuation instructions for the target entity; The target asset data to be appraised is determined according to the asset appraisal instruction; The target asset data to be evaluated is processed using the target base model to obtain the first evaluation data and the target data evaluation requirements. The a specialized intelligent agents process the first evaluation data according to the target data evaluation requirements to obtain the target asset evaluation result.

[0033] Please see Figure 3 , Figure 3 This is a flowchart of an asset valuation method based on multi-agent collaboration provided in an embodiment of this application. The method may include the following steps: S301. The preset asset valuation model is trained using data from the first preset database to obtain the target asset valuation model; the target asset valuation model includes: a target base model and a professional intelligent agents; the a professional intelligent agents correspond to a professional tasks; each professional intelligent agent corresponds to one professional task; a is an integer greater than 1.

[0034] In this embodiment, the preset asset valuation model can be preset in advance or defaulted; each professional task may include one of the following: data quality assessment task, compliance review task, asset value estimation task, risk identification and assessment task, ownership determination task, etc., without limitation.

[0035] It should be explained that the asset valuation method based on multi-agent collaboration provided in this application embodiment can be applied to... Figure 1 The electronic device shown.

[0036] In a specific embodiment, training data can be obtained from a first preset database, and the preset asset valuation model can be trained using the training data to obtain the target asset valuation model.

[0037] Optional, please refer to Figure 4 , Figure 4 This is a schematic diagram of a preset asset valuation model provided in an embodiment of this application. It can be seen that the preset asset valuation model includes: a preset base model and a preset intelligent agents; the preset base model is a general pre-trained large language model; step S301, training the preset asset valuation model using data from a first preset database to obtain a target asset valuation model, includes: A1. Extract the target basic dataset from the first preset database; A2. Train the preset base model using the target basic dataset to obtain the target base model; A3. Based on the a professional tasks, extract a professional training datasets and a professional test datasets from the first preset database; each professional task corresponds to one professional training dataset and one professional test dataset; A4. Using the a professional training datasets and the a professional test datasets, train the a preset intelligent agents to obtain the a professional intelligent agents.

[0038] In this embodiment, the training requirements of the target base model can be clarified first, and the professional task dimensions, format and source requirements of the required data can be determined. Then, the original data that meets the rules can be selected from the first preset database, and invalid and duplicate content can be removed. Next, the selected data can be cleaned, standardized, and labeled, and metadata can be added to adapt to the model training specifications, and finally the target base dataset can be obtained.

[0039] Then, the target base model can be trained using the target base dataset to obtain the target base model. Next, based on 'a' professional tasks, 'a' professional training datasets and 'a' professional test datasets can be extracted from the first pre-set database. Specifically, for each professional task, the original data of the corresponding professional field can be selected from the first pre-set database according to the evaluation dimensions and inference requirements of the professional task. The selected original data is cleaned, deduplicated, standardized, and professionally labeled to form a professional training dataset adapted to the professional task. Independent samples are extracted from the remaining data in the first pre-set database (i.e., data not selected into the professional training dataset) and processed according to the same standards to form a professional test dataset with the same distribution as the professional training dataset. In this way, 'a' professional training datasets and 'a' professional test datasets can be obtained.

[0040] Finally, we can use a professional training dataset and a professional test dataset to train a preset intelligent agents to obtain a professional intelligent agents. Specifically, we will take the first preset intelligent agent and its corresponding first professional training dataset and first professional test dataset as an example. The first preset intelligent agent is any one of the a preset intelligent agents. The specific training process is as follows: Initialization and parameter configuration: Initialize the first preset agent, configure training hyperparameters (e.g., learning rate, batch size, number of iterations, etc.), and load the first professional training dataset.

[0041] Model training: Input the samples from the first professional training dataset into the first preset agent, and iteratively optimize the model parameters through forward propagation, loss calculation and back propagation, so that the first preset agent learns the evaluation logic and reasoning rules of the corresponding professional task.

[0042] Model Validation: After training, the first pre-defined agent is validated using the first professional test dataset to evaluate its performance metrics such as accuracy and recall on professional tasks.

[0043] Parameter tuning and optimization: Adjust hyperparameters or training strategies based on test results, iterate training and validation repeatedly until the performance metrics meet the second preset condition. The second preset condition can be pre-set or defaulted. Specifically, the second preset condition can be: the model training reaches convergence, or the model accuracy is greater than the first preset threshold (e.g., 96%).

[0044] Model solidification: Solidify the trained and optimized model parameters to obtain the first professional intelligent agent with corresponding professional evaluation capabilities.

[0045] Thus, repeating this process a times will yield a specialized intelligent agents.

[0046] It is evident that hierarchical training and specialized data matching maximize model accuracy and evaluation performance. On one hand, the target-based model learns general parsing and requirement decomposition capabilities from the target-based dataset, providing unified foundational support for various specialized agents, avoiding redundant training and improving efficiency. On the other hand, customized training and testing datasets for specific tasks enable each agent to accurately learn domain rules, enhancing professional adaptability and evaluation accuracy. Independent test sets objectively verify generalization ability and stability, reducing evaluation errors.

[0047] Optionally, step A2, training the preset base model using the target base dataset to obtain the target base model, includes: B1. Based on the target base dataset, determine the supervised fine-tuning dataset and the contrastive learning dataset; the contrastive learning dataset includes multiple pairs of positive and negative samples; each pair of positive and negative samples includes an anchor sample, a positive sample, and a negative sample; B2. Supervise the preset base model using the supervised fine-tuning dataset to obtain the first base model; B3. Using the comparative learning dataset, perform comparative learning on the first base model to obtain the second base model; B4. When the second base model meets the first preset condition, the target base model is determined based on the second base model.

[0048] In this embodiment of the application, the first preset condition can be preset in advance or defaulted.

[0049] In a specific embodiment, based on the target base dataset, a supervised fine-tuning dataset and a contrastive learning dataset are determined. Specifically, samples with clearly labeled evaluation results (e.g., data compliance conclusions, asset value ranges, asset risks, etc.) can be selected from the target base dataset to directly constitute the supervised fine-tuning dataset, which is used to guide the target base model in learning the basic rules and standard output paradigms for data asset evaluation. Additionally, based on the original samples in the target base dataset, positive and negative samples with semantically similar but different evaluation attributes can be generated through data augmentation, attribute replacement, and scene perturbation. An original sample and its corresponding positive and negative samples form a positive-negative sample pair. In this way, multiple positive-negative sample pairs can be obtained, which constitute the contrastive learning dataset. Anchor samples are the original samples; positive samples are similar samples with the same evaluation attributes as the original samples (i.e., samples belonging to the same category / having the same evaluation conclusion as the original samples); negative samples are dissimilar samples with contradictory evaluation attributes as the original samples (i.e., samples belonging to different categories / having different evaluation conclusions than the original samples). These are used to guide the target base model in learning the correlation between data features and evaluation results, improving the model's ability to discern subtle differences in the data.

[0050] Next, the pre-defined base model can be supervisedly fine-tuned using the supervised fine-tuning dataset. The specific process is as follows: Data preparation and input: Input the samples (including input data and their corresponding labeled evaluation results) from the supervised fine-tuning dataset into the preset base model in batches to provide the model with clear supervision signals.

[0051] Forward propagation and prediction output: The pre-defined base model extracts features and infers from the input data to generate corresponding evaluation and prediction results.

[0052] Loss calculation: The model's prediction results are compared with the labeled results in the supervised fine-tuning dataset. The prediction error is calculated using a preset loss function (e.g., cross-entropy loss, mean squared error loss) to obtain the loss value.

[0053] Backpropagation and parameter optimization: Based on the loss value, the gradient of the model parameters is calculated through the backpropagation algorithm, and an optimizer (e.g., Adam optimizer) is used to iteratively update the parameters of the preset base model, gradually reducing the prediction error.

[0054] Iterative training and convergence: Repeat the forward propagation, loss calculation and backpropagation process until the model's performance metrics (e.g. accuracy, loss value) on the validation set reach the preset convergence conditions, complete supervised fine-tuning, and obtain the first base model with basic data asset evaluation capabilities.

[0055] Then, the first base model can be compared and learned using a contrastive learning dataset, as follows: Sample pair input and feature encoding: Input the positive and negative sample pairs (anchor sample, positive sample, negative sample) in the contrastive learning dataset into the first base model, extract the feature vector of each sample, and obtain the corresponding feature representation.

[0056] Similarity calculation and loss construction: Calculate the similarity between the original sample (i.e., anchor sample) and the feature vector of the positive sample, as well as the similarity between the original sample and the feature vector of each negative sample; based on the contrastive loss function (e.g., InfoNCE loss function), construct the contrastive loss value for model training with the goal of bringing similar sample features closer together and distancing dissimilar sample features further apart.

[0057] Backpropagation and parameter optimization: Based on the comparative loss value, the gradient of the model encoder parameters is calculated through the backpropagation algorithm. The optimizer is used to iteratively update the parameters of the first base model, thereby enhancing the model's ability to discriminate subtle differences in data features.

[0058] Iterative training and convergence: Repeated sample encoding, similarity calculation, loss construction and parameter optimization process until the model's feature discrimination and convergence index on the validation set reach the convergence condition, completing comparative learning and obtaining a second base model with better feature extraction and discrimination capabilities.

[0059] When the second base model meets the first preset condition, the target base model is determined based on the second base model. Specifically, the first preset condition may be: the model training reaches a convergence state, or the model accuracy is greater than the second preset threshold (e.g., 98%).

[0060] If the second base model does not meet the first preset condition, new training data is acquired and the model is trained again until it meets the first preset condition, at which point it is identified as the target base model.

[0061] Thus, by training the base model using a combination of supervised fine-tuning and contrastive learning, supervised fine-tuning first allows the model to quickly master the basic evaluation rules, and then contrastive learning improves its ability to discriminate data features. This complementary two-stage training ensures both standardized evaluation and enhanced model robustness. Finally, the target base model is determined according to preset conditions to ensure performance targets are met and to improve evaluation accuracy and stability.

[0062] S302, Obtain asset valuation instructions for the target object.

[0063] In this application embodiment, the target object can be one of the following: an individual, a school, a company, a hospital, etc., without limitation.

[0064] In a specific embodiment, an asset appraisal request initiated by the target object can be received, and key information can be extracted from the asset appraisal request, including the appraisal type (e.g., value appraisal, risk appraisal, quality appraisal), appraisal dimensions, appraisal standards and output requirements. The parsed key information can be transformed into executable structured instructions, i.e., asset appraisal instructions.

[0065] S303. Determine the target asset data to be appraised according to the asset appraisal instruction.

[0066] In this embodiment of the application, data can be collected according to the asset appraisal instruction to obtain the target asset data to be appraised.

[0067] Optionally, the asset valuation instruction includes: the asset valuation period and information on the assets to be valued; step S303, determining the target asset data to be valued according to the asset valuation instruction, includes: S31. Determine the reference asset data corresponding to the target object based on the asset valuation period; S32. Determine the target asset data to be evaluated based on the asset information to be evaluated and the reference asset data.

[0068] In this embodiment, all assets owned by the target object during the asset valuation period can be collected to obtain reference asset data. Then, key asset characteristics can be extracted from the asset information to be valued, such as asset name, asset size, industry of the asset, geographical scope, etc., which are not limited here. Next, corresponding asset data is extracted from the reference asset data according to the key asset characteristics, thereby obtaining the target asset data to be valued. For example, assuming that the key asset characteristic extracted from the asset information to be valued is "xx asset in xx industry", the corresponding asset data, i.e., the target asset data to be valued, can be extracted from the reference asset data according to the key asset characteristics.

[0069] In this way, by first filtering reference asset data by time period, the timeliness of the data is ensured to match the assessment scenario, avoiding assessment distortion due to time deviation; then, by combining the information of the asset to be assessed with the reference data to determine the target data, the precise matching of features is achieved, ensuring that the data is highly relevant to the assessment object, effectively improving the accuracy and rationality of the assessment.

[0070] S304. The target asset data to be evaluated is processed through the target base model to obtain the first evaluation data and the target data evaluation requirements.

[0071] In this embodiment of the application, the target asset data to be evaluated can be input into the target base model to obtain the first evaluation data and the target data evaluation requirements.

[0072] Optionally, the target foundation model includes: a data evaluation module and a demand evaluation module; the demand evaluation module is equipped with preset evaluation rules; step S304, processing the target asset data to be evaluated through the target foundation model to obtain first evaluation data and target data evaluation requirements, includes: S41. The data evaluation module processes the target asset data to be evaluated to obtain the first evaluation data; S42. Through the demand assessment module, determine the assessable dimensions in the first assessment data according to the preset assessment rules to obtain at least one assessable dimension; determine the target data assessment demand according to the asset assessment instruction and the at least one assessable dimension.

[0073] In this embodiment of the application, the preset evaluation rules can be preset in advance or defaulted.

[0074] In a specific embodiment, the target asset data to be evaluated can be input into the data evaluation module. The data evaluation module analyzes, calculates, and extracts features based on the model's own algorithm or logic, and outputs preliminary evaluation results, namely the first evaluation data. Alternatively, the first evaluation data can be input into the demand evaluation module. The demand evaluation module analyzes the first evaluation data according to preset evaluation rules to obtain at least one evaluable dimension. For example, let's assume the preset evaluation rules are: Rule 1: When appraising corporate assets, "revenue" and "net profit" must be assessed.

[0075] Rule 2: Technology companies need to have their "number of patents" additionally assessed.

[0076] Rule 3: "Number of employees" and "Years of establishment" are non-core dimensions. Unless the user's assessment instructions explicitly require an assessment, these two dimensions will not be considered in the enterprise asset assessment.

[0077] Assume the first set of assessment data includes: revenue data, net profit data, employee data, patent data, and years of establishment data; the demand assessment module compares each item in the first set of assessment data with the preset assessment rules one by one.

[0078] Matching rule 1: Determine "revenue" and "net profit" as evaluable dimensions.

[0079] Matching rule 2: Determine "number of patents" as an evaluable dimension.

[0080] Matching rule 3: Exclude "number of employees" and "years since establishment".

[0081] Ultimately, three assessable dimensions are obtained: revenue, net profit, and number of patents. Finally, based on the asset valuation instruction and at least one assessable dimension, the target data assessment requirements can be determined. Specifically, the asset valuation instruction and at least one assessable dimension can be integrated to obtain the target data assessment requirements. For example, assuming the asset valuation instruction is "assess the investment value of xx company in 2025, output a comprehensive score and core dimension analysis, focusing on profitability and technological strength," and at least one assessable dimension is "revenue, net profit, and number of patents," integrating these two results in the target data assessment requirements being "calculate the comprehensive investment value score of xx company in 2025 based on the three dimensions of revenue, net profit, and number of patents, with a weighting of 4:3:3, and output the scores and analysis for each dimension, focusing on the impact of profitability and technological strength on investment value."

[0082] In this way, the first evaluation data is obtained through the data evaluation module, and then the evaluation dimensions are selected according to the rules by the demand evaluation module. Finally, the target demand is determined in combination with the instructions, forming a closed loop of data processing, dimension selection and demand matching, which improves the pertinence, rationality and efficiency of the evaluation and ensures that the results meet the requirements of the instructions.

[0083] S305. The a professional intelligent agents process the first evaluation data according to the target data evaluation requirements to obtain the target asset evaluation result.

[0084] Optionally, step S305, whereby the a professional intelligent agents process the first evaluation data according to the target data evaluation requirements to obtain the target asset evaluation result, includes: C1. Select the professional intelligent agents that correspond to the target data evaluation requirements from the a professional intelligent agents to obtain b professional intelligent agents; b is a positive integer less than or equal to a. C2. Using RAG technology, retrieve specialized knowledge from a second preset database for the b specialized intelligent agents to obtain b specialized knowledge bases; each specialized knowledge base corresponds to one specialized intelligent agent; the second preset database includes various professional knowledge data; C3. The b professional intelligent agents evaluate the first evaluation data according to the preset prompting strategy and the b exclusive knowledge bases to obtain b asset evaluation results. C4. Determine the target asset valuation result based on the valuation results of the b assets.

[0085] In this embodiment of the application, both the second preset database and the preset prompting strategy can be preset in advance or defaulted; wherein, the preset prompting strategy may include one of the following: structured multi-stage prompting strategy, role-based prompting strategy, iterative optimization prompting strategy, etc., which are not limited here.

[0086] In a specific embodiment, professional intelligent agents corresponding to the target data evaluation requirements can be selected from a number of professional intelligent agents to obtain b professional intelligent agents. Specifically, the target data evaluation requirements can be parsed and broken down into executable professional tasks to obtain b professional tasks. For example, assuming the target data evaluation requirement is "enterprise investment value evaluation", it can be broken down into three professional tasks: "asset value estimation task", "compliance review task", and "risk identification and assessment task", i.e., b=3. Then, b professional intelligent agents corresponding to these b professional tasks can be selected from a number of professional intelligent agents.

[0087] Furthermore, RAG technology can be used to retrieve specialized knowledge from a second pre-set database for b specialized agents, resulting in b specialized knowledge bases. Specifically, for each of the b specialized agents, the specialized task corresponding to that agent is first determined. Then, based on the specialized task, corresponding core search terms and professional domain keywords are determined. Using the core search terms and professional domain keywords as search conditions, RAG technology is used to retrieve knowledge fragments such as text, data, rules, and cases highly related to the specialized task from the second pre-set database, obtaining search results. The search results are then deduplicated and filtered, removing irrelevant or low-quality information and retaining the specialized knowledge most closely matching the agent's specialized task. The filtered specialized knowledge is then vectorized and stored to form an independent specialized knowledge base for that agent, which can be called upon during evaluation and reasoning. In this way, b specialized knowledge bases can be obtained.

[0088] Furthermore, b specialized intelligent agents can evaluate the first evaluation data based on preset prompting strategies and b exclusive knowledge bases to obtain b asset evaluation results. Finally, the target asset evaluation result can be determined based on the b asset evaluation results. Specifically, the b asset evaluation results can be deduplicated and merged to obtain the target asset evaluation result.

[0089] In this way, by selecting specialized intelligent agents on demand and using RAG technology to build a dedicated knowledge base for them, the precise matching of knowledge and tasks can be achieved. Each intelligent agent independently evaluates itself by combining preset prompting strategies and dedicated knowledge, which effectively improves the professionalism and accuracy of the evaluation results. Finally, by integrating the evaluation results of multiple intelligent agents, the bias of a single evaluation is effectively avoided, thereby improving the reliability of the overall evaluation.

[0090] In some embodiments, please refer to Figure 5 , Figure 5 This application provides a flowchart for generating a data asset valuation report, as detailed below: 1. Input phase: Input data asset fields and documents: Users can manually upload metadata fields of the data assets to be evaluated (e.g., asset type, ownership information) and original business documents (e.g., compliance statements, technical documents, asset information documents, etc.) to provide raw data for all subsequent analysis modules. It is the starting point of the entire process.

[0091] 2. Parallel preprocessing and knowledge preparation stage: Input information is simultaneously distributed to three modules, where basic processing and knowledge support work are carried out in parallel: (1) Multidimensional data asset knowledge base: It can retrieve authoritative background knowledge such as industry standards, assessment rules, and compliance requirements that match the data assets to be assessed from a pre-set multidimensional data asset knowledge base, providing a unified knowledge benchmark for subsequent diagnostic analysis and ensuring that the assessment logic conforms to industry norms.

[0092] (2) Hypertext Transfer Protocol Service (i.e., HTTP service): By calling external interfaces (such as regulatory databases and third-party compliance platforms), external verification information of the data assets to be evaluated can be obtained, including compliance status and related risk events; supplementing external authoritative data can enhance the comprehensiveness and credibility of the evaluation.

[0093] (3) Target base model: The target base model parses, semantically understands, and structures the input raw documents to extract key information and core logic; it transforms unstructured raw documents into structured data that can be recognized by subsequent modules, serving as the foundational computing engine for the entire process.

[0094] 3. Diagnostic Analysis Phase: (1) Document Summary: Based on the output of the target base model, the core content of the document is extracted to generate a structured document summary, providing efficient input for subsequent agent analysis.

[0095] (2) Agent Engine: The intelligent agent engine (comprising a specialized intelligent agents) integrates document summarization, rules from a multi-dimensional data asset knowledge base, and external data from HTTP services. Through pre-defined reasoning logic, it performs diagnostic analysis to identify problems and risks in data assets. It is a core analysis module for asset valuation, driving the reasoning and decision-making throughout the diagnostic process.

[0096] (3) Diagnostic summary: The analysis results from the intelligent agent engine are summarized to generate structured diagnostic conclusions that include the current status, problems, and risks of data assets, providing a conclusive basis for subsequent rectification and reporting.

[0097] (4) Regularization removes redundancy: The diagnostic summary is cleaned to remove redundant logic tags, intermediate reasoning traces, and other non-user-friendly content (i.e., think, which refers to intermediate thinking traces, logic drafts, or debugging tags generated by large models or agents during the reasoning process, and is not effective content for end users), so as to standardize the format and content of the diagnostic summary and make it more in line with the readability requirements of the final report.

[0098] (5) Data Asset Rectification Knowledge Base: The system receives the regularized diagnostic summary and matches it with the corresponding rectification rules, best practices, and industry cases in the data asset rectification knowledge base, providing actionable rules and references for generating rectification opinions.

[0099] 4. Rectification Output and Report Generation Stage: (1) Rectification suggestions: The target foundation model combines diagnostic conclusions with the rules of the rectification knowledge base to output specific and actionable rectification suggestions for data asset issues.

[0100] (2) Format processing: Standardize the formatting of diagnostic summaries and rectification suggestions, including fonts, layouts, and numbering, to ensure the professionalism and standardization of the final report and improve document readability and delivery experience.

[0101] (3) Data asset valuation report: The integrated formatted diagnostic summary and rectification suggestions are combined to generate a complete data asset assessment report, providing users with deliverables that include diagnostic conclusions and rectification plans.

[0102] In some embodiments, please refer to Figure 6 , Figure 6 This is a flowchart of a data asset assessment provided in an embodiment of this application, as detailed below: 1. Field Input: The process begins here, where the user enters the name of the company to be evaluated (e.g., the companyName field) as the initial data for the entire process.

[0103] 2. Hypertext Transfer Protocol Service: After receiving the company name, the system can call the backend HTTP service and use code scripts to simulate a human login to the company credit information website. Its core function is compliance review and information crawling. It will automatically query the company's penalty records, negative events, and other credit-related information to obtain the company's credit data.

[0104] 3. Specialized intelligent agents: The company's credit data will be input into a specialized intelligent agent (i.e., the aforementioned a specialized intelligent agent). The agent will analyze the data based on authoritative information sources and generate an accurate assessment of the company's credit and compliance status, i.e., the credit assessment result.

[0105] 4. Digital Asset Valuation: This is the end of the process, obtaining the credit assessment results output by the specialized intelligent agent, using it as one of the core bases to conduct a complete digital asset assessment, and finally outputting a comprehensive assessment conclusion. Specifically, the asset assessment method based on multi-agent collaboration provided in the embodiments of this application can be executed to conduct a digital asset assessment of the enterprise to be assessed.

[0106] Optionally, the preset prompting strategy includes a structured multi-stage prompting strategy; step C3, whereby the b professional intelligent agents process the first evaluation data according to the preset prompting strategy and the b exclusive knowledge bases to obtain b asset evaluation results, includes: D1. Obtain the first professional intelligent agent and its corresponding first exclusive knowledge base; the first professional intelligent agent is any one of the b professional intelligent agents. D2. Based on the first dedicated knowledge base, determine the reasoning hard constraints corresponding to the first evaluation data; D3. Based on the structured multi-stage prompting strategy, determine the multiple inference stages and multiple inference stage constraints corresponding to the first professional intelligent agent; D4. The first professional intelligent agent processes the first evaluation data according to the multiple inference stages, the multiple inference stage constraints, and the hard inference constraints to obtain the asset evaluation result corresponding to the first professional intelligent agent.

[0107] In this embodiment, a first professional intelligent agent and its corresponding first dedicated knowledge base can be obtained first. Then, based on the first dedicated knowledge base, the hard constraints of inference corresponding to the first evaluation data can be determined. Specifically, based on at least one evaluable dimension, corresponding industry standards, regulatory requirements, evaluation criteria, threshold ranges, and other restrictive knowledge can be retrieved from the first dedicated knowledge base to obtain multiple restrictive knowledge, such as revenue growth rate ≥ 5%, risk level not higher than B, etc., without limitation. Then, rules irrelevant to the current evaluation task are filtered out from these multiple restrictive knowledge, and hard constraints that directly constrain the evaluation logic, calculation method, or result range are retained, i.e., hard constraints of inference.

[0108] Next, based on the structured multi-stage prompting strategy, multiple inference stages and constraints corresponding to the first professional agent can be determined. Specifically, according to the asset appraisal task logic, the complete inference process is broken down into multiple inference stages, such as the knowledge retrieval stage, analysis and calculation stage, conclusion derivation stage, and report generation stage. Then, structured output requirements and anti-illusion mechanisms can be embedded in each inference stage to obtain multiple inference stage constraints, such as: During the knowledge retrieval stage, the corresponding reasoning stage is constrained to: force the output of authoritative knowledge sources retrieved, and answer questions based solely on the content of the dedicated knowledge base; During the analysis and calculation phase, the corresponding reasoning phase is constrained by the following: presenting dimensional data and calculation logic in a fixed format, and prohibiting unfounded speculation. During the conclusion derivation stage, the corresponding reasoning stage constraint is: to explicitly mark conflicting conclusions to ensure that the logical chain is traceable; During the report generation phase, the corresponding reasoning phase constraints are: follow the standard template for output to ensure consistent format and accurate semantics.

[0109] Finally, the first professional intelligent agent can process the first assessment data according to multiple inference stages, multiple inference stage constraints, and hard inference constraints to obtain the asset assessment result corresponding to the first professional intelligent agent. Specifically, the first professional intelligent agent executes step by step according to these multiple inference stages, and each stage strictly adheres to the corresponding stage constraints (e.g., structured output, anti-illusion mechanism), while following the hard inference constraints (e.g., industry standards, regulatory requirements) throughout the process to analyze, calculate, and verify the first assessment data, and finally generate the asset assessment result corresponding to the first professional intelligent agent that conforms to the specifications and is accurate and reliable.

[0110] In summary, the asset appraisal method based on multi-agent collaboration described in this application obtains a target asset appraisal model through model training. This target asset appraisal model includes a target base model and a specialized intelligent agents. The target base model quickly completes the parsing of the asset data to be appraised and the decomposition of appraisal requirements. Then, the a specialized intelligent agents process the first appraisal data in a professional and parallel manner according to the appraisal requirements. This achieves full-process automation and collaboration from data processing and requirements analysis to multi-dimensional appraisal, replacing inefficient links such as experience dependence, serial processing, and report writing in traditional manual appraisal, thereby improving the efficiency of asset appraisal.

[0111] Please see Figure 7 , Figure 7 This is a functional block diagram of an asset appraisal device based on multi-agent collaboration provided in an embodiment of this application. The asset appraisal device 700 based on multi-agent collaboration includes: a training module 701, an acquisition module 702, and an asset appraisal module 703, wherein: The training module 701 is used to train a preset asset valuation model using data from a first preset database to obtain a target asset valuation model. The target asset valuation model includes: a target base model and a professional intelligent agents. The a professional intelligent agents correspond to a professional tasks. Each professional intelligent agent corresponds to one professional task. a is an integer greater than 1. The acquisition module 702 is used to acquire the asset valuation instructions of the target object; The asset valuation module 703 is used to determine the target asset data to be valued according to the asset valuation instruction; process the target asset data to be valued through the target base model to obtain the first valuation data and the target data valuation requirements; and process the first valuation data according to the target data valuation requirements through the a professional intelligent agents to obtain the target asset valuation result.

[0112] Optionally, the preset asset valuation model includes: a preset base model and a preset intelligent agents; the preset base model is a general pre-trained large language model; in terms of training the preset asset valuation model using data from the first preset database to obtain the target asset valuation model, the training module 701 is specifically used for: Extract the target basic dataset from the first preset database; The target base model is obtained by training the preset base model using the target basic dataset; Based on the aforementioned a professional tasks, a professional training datasets and a professional test datasets are extracted from the first preset database; each professional task corresponds to one professional training dataset and one professional test dataset; Using the a professional training datasets and the a professional test datasets, the a preset intelligent agents are trained to obtain the a professional intelligent agents.

[0113] Optionally, in the step of training the preset base model using the target base dataset to obtain the target base model, the training module 701 is specifically used for: Based on the target base dataset, a supervised fine-tuning dataset and a contrastive learning dataset are determined; the contrastive learning dataset includes multiple pairs of positive and negative samples; each pair of positive and negative samples includes an anchor sample, a positive sample, and a negative sample. The preset base model is subjected to supervised fine-tuning using the supervised fine-tuning dataset to obtain the first base model; The first base model is subjected to comparative learning using the aforementioned comparative learning dataset to obtain the second base model; When the second base model meets the first preset condition, the target base model is determined based on the second base model.

[0114] Optionally, the target foundation model includes: a data evaluation module and a demand evaluation module; the demand evaluation module is configured with preset evaluation rules; in terms of processing the target asset data to be evaluated through the target foundation model to obtain the first evaluation data and the target data evaluation demand, the asset evaluation module 703 is specifically used for: The data evaluation module processes the target asset data to be evaluated to obtain the first evaluation data; The demand assessment module determines the assessable dimensions in the first assessment data according to the preset assessment rules, thereby obtaining at least one assessable dimension; and determines the target data assessment demand according to the asset assessment instruction and the at least one assessable dimension.

[0115] Optionally, in the process of processing the first evaluation data according to the target data evaluation requirements by the a professional intelligent agents to obtain the target asset evaluation result, the asset evaluation module 703 is specifically used for: From the a professional intelligent agents, select the professional intelligent agents that correspond to the target data evaluation requirements to obtain b professional intelligent agents; b is a positive integer less than or equal to a. Using RAG technology, specific knowledge is retrieved from a second preset database for the b specialized intelligent agents, resulting in b specialized knowledge bases; each specialized knowledge base corresponds to one specialized intelligent agent; the second preset database includes various professional knowledge data. The b specialized intelligent agents evaluate the first evaluation data according to the preset prompting strategy and the b exclusive knowledge bases to obtain b asset evaluation results; The valuation result of the target asset is determined based on the valuation results of the b assets.

[0116] Optionally, the preset prompting strategy includes a structured multi-stage prompting strategy; in the process of processing the first evaluation data by the b professional intelligent agents according to the preset prompting strategy and the b exclusive knowledge bases to obtain b asset evaluation results, the asset evaluation module 703 is specifically used for: Obtain the first professional intelligent agent and its corresponding first exclusive knowledge base; the first professional intelligent agent is any one of the b professional intelligent agents; Based on the first dedicated knowledge base, determine the reasoning hard constraints corresponding to the first evaluation data; Based on the structured multi-stage prompting strategy, multiple inference stages and multiple inference stage constraints corresponding to the first professional intelligent agent are determined; The first professional intelligent agent processes the first evaluation data according to the multiple inference stages, the multiple inference stage constraints, and the hard inference constraints to obtain the asset evaluation result corresponding to the first professional intelligent agent.

[0117] Optionally, the asset valuation instruction includes: the asset valuation period and information on the assets to be valued; in determining the target asset data to be valued according to the asset valuation instruction, the asset valuation module 703 is specifically used for: Based on the asset valuation period, determine the reference asset data corresponding to the target object; Based on the asset information to be evaluated and the reference asset data, the target asset data to be evaluated is determined.

[0118] In specific implementations, the asset appraisal device 700 based on multi-agent collaboration described in the embodiments of the present invention can also execute other implementations described in the asset appraisal method based on multi-agent collaboration provided in the embodiments of the present invention, which will not be repeated here.

[0119] Please see Figure 8 , Figure 8 This is a schematic diagram of another electronic device provided in an embodiment of this application. The electronic device may include a processor, a memory, a communication interface, and one or more programs. The processor, memory, and communication interface can be interconnected via a bus. The one or more programs are stored in the memory and configured to be executed by the processor. In this embodiment, the programs include instructions for performing some or all of the steps in the method embodiments described above.

[0120] This application also provides a computer-readable storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the methods described in the above method embodiments, wherein the computer includes an electronic device.

[0121] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. The computer program product may be a software installation package, and the computer may include an electronic device.

[0122] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0123] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0125] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0126] The steps of the methods or algorithms described in the embodiments of this application can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (EEPROM), registers, hard disk, portable hard disk, read-only optical disk (CD-ROM), or any other form of storage medium well known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Furthermore, the ASIC can reside in a terminal device or management device. Alternatively, the processor and storage medium can exist as discrete components in the terminal device or management device.

[0127] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in the embodiments of this application can be implemented, in whole or in part, by software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0128] The aforementioned computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media.

[0129] The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0130] The modules / units included in the various devices and products described in the above embodiments can be software modules / units, hardware modules / units, or a combination of both. For example, for devices and products applied to or integrated into a chip, all modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits. For devices and products applied to or integrated into a chip module, all modules / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The implementation is achieved through a software program that runs on the processor integrated within the chip module. The remaining modules / units (if any) can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into terminal equipment, each of their modules / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components within the terminal equipment. Alternatively, at least some modules / units can be implemented through a software program that runs on the processor integrated within the terminal equipment, while the remaining modules / units (if any) can be implemented using hardware methods such as circuits.

[0131] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the embodiments of this application. It should be understood that the above descriptions are merely specific embodiments of the embodiments of this application and are not intended to limit the protection scope of the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solutions of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. An asset valuation method based on multi-agent collaboration, characterized in that, include: The target asset valuation model is obtained by training the preset asset valuation model using data from the first preset database. The target asset evaluation model includes: a target foundation model and a professional intelligent agents; the a professional intelligent agents correspond to a professional tasks; each professional intelligent agent corresponds to one professional task; a is an integer greater than 1; Obtain asset valuation instructions for the target entity; The target asset data to be appraised is determined according to the asset appraisal instruction; The target asset data to be evaluated is processed using the target base model to obtain the first evaluation data and the target data evaluation requirements. The a specialized intelligent agents process the first evaluation data according to the target data evaluation requirements to obtain the target asset evaluation result.

2. The method as described in claim 1, characterized in that, The preset asset evaluation model includes: a preset base model and a preset intelligent agents; the preset base model is a general pre-trained large language model. The step of training a preset asset valuation model using data from a first preset database to obtain a target asset valuation model includes: Extract the target basic dataset from the first preset database; The target base model is obtained by training the preset base model using the target basic dataset; Based on the aforementioned a professional tasks, a professional training datasets and a professional test datasets are extracted from the first preset database; each professional task corresponds to one professional training dataset and one professional test dataset; Using the a professional training datasets and the a professional test datasets, the a preset intelligent agents are trained to obtain the a professional intelligent agents.

3. The method as described in claim 2, characterized in that, The step of training the preset base model using the target basic dataset to obtain the target base model includes: Based on the target base dataset, a supervised fine-tuning dataset and a contrastive learning dataset are determined; the contrastive learning dataset includes multiple pairs of positive and negative samples; each pair of positive and negative samples includes an anchor sample, a positive sample, and a negative sample. The preset base model is subjected to supervised fine-tuning using the supervised fine-tuning dataset to obtain the first base model; The first base model is subjected to comparative learning using the aforementioned comparative learning dataset to obtain the second base model; When the second base model meets the first preset condition, the target base model is determined based on the second base model.

4. The method according to any one of claims 1-3, characterized in that, The target foundation model includes: a data evaluation module and a demand evaluation module; the demand evaluation module is equipped with preset evaluation rules. The process of processing the target asset data to be evaluated through the target base model to obtain the first evaluation data and target data evaluation requirements includes: The data evaluation module processes the target asset data to be evaluated to obtain the first evaluation data; The demand assessment module determines the assessable dimensions in the first assessment data according to the preset assessment rules, thereby obtaining at least one assessable dimension; and determines the target data assessment demand according to the asset assessment instruction and the at least one assessable dimension.

5. The method according to any one of claims 1-3, characterized in that, The step of processing the first assessment data based on the target data assessment requirements by the a professional intelligent agents to obtain the target asset assessment result includes: From the a professional intelligent agents, select the professional intelligent agents that correspond to the target data evaluation requirements to obtain b professional intelligent agents; b is a positive integer less than or equal to a. Using RAG technology, specific knowledge is retrieved from a second preset database for the b specialized intelligent agents, resulting in b specialized knowledge bases; each specialized knowledge base corresponds to one specialized intelligent agent; the second preset database includes various professional knowledge data. The b specialized intelligent agents evaluate the first evaluation data according to the preset prompting strategy and the b exclusive knowledge bases to obtain b asset evaluation results; The valuation result of the target asset is determined based on the valuation results of the b assets.

6. The method as described in claim 5, characterized in that, The preset prompting strategy includes a structured multi-stage prompting strategy; The process involves the b specialized intelligent agents processing the first assessment data according to a preset prompting strategy and the b dedicated knowledge bases to obtain b asset assessment results, including: Obtain the first professional intelligent agent and its corresponding first exclusive knowledge base; the first professional intelligent agent is any one of the b professional intelligent agents; Based on the first dedicated knowledge base, determine the reasoning hard constraints corresponding to the first evaluation data; Based on the structured multi-stage prompting strategy, multiple inference stages and multiple inference stage constraints corresponding to the first professional intelligent agent are determined; The first professional intelligent agent processes the first evaluation data according to the multiple inference stages, the multiple inference stage constraints, and the hard inference constraints to obtain the asset evaluation result corresponding to the first professional intelligent agent.

7. The method according to any one of claims 1-3, characterized in that, The asset valuation instruction includes: the asset valuation period and information on the assets to be valued; The step of determining the target asset data to be appraised according to the asset appraisal instruction includes: Based on the asset valuation period, determine the reference asset data corresponding to the target object; Based on the asset information to be evaluated and the reference asset data, the target asset data to be evaluated is determined.

8. An asset appraisal device based on multi-agent collaboration, characterized in that, The device includes: a training module, an acquisition module, and an asset evaluation module, wherein: The training module is used to train a preset asset valuation model using data from a first preset database to obtain a target asset valuation model. The target asset valuation model includes: a target base model and a professional intelligent agents. The a professional intelligent agents correspond to a professional tasks. Each professional intelligent agent corresponds to one professional task. a is an integer greater than 1. The acquisition module is used to acquire the asset valuation instructions of the target object; The asset valuation module is used to determine the target asset data to be valued according to the asset valuation instruction; process the target asset data to be valued through the target base model to obtain the first valuation data and the target data valuation requirements; and process the first valuation data according to the target data valuation requirements through the a professional intelligent agents to obtain the target asset valuation result.

9. An electronic device, characterized in that, include: Processor, memory, communication interface, and one or more programs; The one or more programs are stored in the memory and configured to be executed by the processor, the programs including instructions for performing the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program for storing electronic data interchange, wherein the computer program causes a computer to perform the method as described in any one of claims 1-7.

Images